Revenue Equivalence, Winner's Curse, and VCG vs GSP in Auction Theory

The expected payment of a bidder with valuation (V_i) equals the expected payment at the lowest valuation plus the integral of the derivative of the allocation probability. This formula holds only under strong assumptions: valuations must be independent, drawn from an interval, and the allocation probability must be differentiable. If values lie outside the interval or independence fails, the revenue‑equivalence result collapses.

The Winner’s Curse

The winner’s curse describes a situation where the winner of an auction pays more than the item’s true value because the winner overestimates that value. In common‑value settings—such as wallet auctions, oil‑drilling rights, government procurement, or painting contracts—each bidder forms a noisy estimate of the same underlying value. The highest estimate is statistically likely to be an overestimate, so the winning bidder ends up overpaying. Paul Milgrom’s work shows that when valuations are not private, the curse becomes a systematic risk for participants.

Sponsored Search Auctions

Search engines run keyword‑specific auctions for multiple heterogeneous slots. Each slot (j) has a click‑through rate (\alpha_j), and each advertiser (i) has a value per click (V_i). The Generalized Second Price (GSP) auction allocates slots to the highest bidders, charging each winner the next highest bid. Because losing the top slot costs less than in a single‑item auction, bidders shade their bids, creating instability in real‑time bidding environments. Google’s model exemplifies this structure, and firms such as Uber and Lyft regularly compete in these auctions.

GSP Payment

For slot (k), the payment is simply the next highest bid: [ p_k = \text{next highest bid}. ]

Vickrey‑Clarke‑Groves (VCG) Mechanism

The VCG mechanism generalizes the second‑price auction by charging each bidder for the externality it imposes on others. The mechanism collects reported valuation functions (\hat v_i), selects the allocation (x^) that maximizes total reported welfare, and computes each payment as: [ \text{Payment}i = \bigl(\max{x_{-i}} \sum_{j\neq i} \hat v_j(x_{-i})\bigr) - \bigl(\sum_{j\neq i} \hat v_j(x^)\bigr). ] This payment equals the loss in welfare that other bidders suffer because bidder (i) participates. Truthful reporting is a dominant strategy, making VCG strategy‑proof regardless of others’ bids.

VCG Payment for Sponsored Search

For slot (j), the VCG payment aggregates lower bids weighted by differences in click‑through rates: [ p_j = \sum_{k=j+1}^{n} b_k \frac{\alpha_{k-1} - \alpha_k}{\alpha_j}. ] The weights sum to one when there are fewer slots than bidders, ensuring the payment reflects the marginal contribution of each higher slot.

Comparing GSP and VCG

GSP charges the next highest bid, while VCG charges a weighted average of all lower bids. The VCG formula internalizes the externality each bidder imposes, eliminating the incentive to shade bids. In contrast, GSP’s simpler rule creates opportunities for strategic underbidding, leading to less efficient outcomes and potential instability.

Key Takeaways

• Revenue equivalence requires independent private values, interval‑bounded valuations, and differentiable allocation probabilities; violating any assumption invalidates the formula.
• The winner’s curse arises when bidders overestimate a common value, causing the winner to pay more than the item’s true worth.
• Sponsored search auctions use GSP, which allocates slots to highest bidders but encourages bid shading because payments depend only on the next highest bid.
• The VCG mechanism charges each bidder for the externality it creates, guaranteeing strategy‑proofness and truthful bidding as a dominant strategy.
• VCG payments in keyword auctions weight all lower bids by click‑through‑rate differences, producing a more efficient outcome than GSP’s single‑bid reference.